Discrimination of Human Lung Neoplasm from Normal
Lung by Two Target Genes
Hans-Stefan Hofmann, Gesine Hansen, Stefan Burdach, Babett Bartling, Rolf-Edgar Silber, and Andreas Simm
Department of Cardiothoracic Surgery, Department of Pediatrics, and Children’s Cancer Research Center, Martin Luther University
Halle-Wittenberg, Halle; and Department of Pediatrics, Technical University of Munich, Munich, Germany
Simple tools for discrimination of lung tissues can be useful in a fast
machine-aided diagnosis, for example, by tumor-specific microarrays.
We demonstrate that an easy ratio technique, based on the expression levels of only two genes differentially expressed in lung tumor
and normal lung samples, allows discrimination of normal and neoplastic lung with a sensitivity of 100% and specificity of 90.5%. DNA
microarray analysis of 99 lung tumor samples and 15 normal lung
tissues revealed that receptor for advanced glycation end products
(RAGE) mRNA is reduced fourfold (p ⫽ 7.8 ⫻ 10⫺11) and cyclin-B2
mRNA is upregulated twofold (p ⫽ 5.9 ⫻ 10⫺18) in lung carcinoma
compared with normal lung. The microarray-calculated expression
ratio of RAGE to cyclin-B2 was used in polymerase chain reaction
analysis of 84 independent blinded samples to discriminate tumor
and corresponding normal lung tissues. In 94.7% of the samples
this quotient correctly distinguished non–small cell lung cancer from
normal lung tissue, suggesting the RAGE/cyclin-B2 quotient as a
potential means for diagnosis of lung cancer.
Keywords: cyclin-B2; lung cancer; lung metastases; receptor for
advanced glycation end products; tissue discrimination
Lung cancer is the leading cause of death worldwide among
patients with cancer, with non–small cell lung cancer (NSCLC)
accounting for about 80% of newly diagnosed cases (1). Current
5-year lung cancer survival rates are estimated at 14% (2). A
patient’s survival depends on histology and cancer stage. The
current lung cancer classification is based on clinicopathological
features. However, these methods are insufficient to reflect the
complicated underlying molecular events that drive the neoplastic process (3). After sequencing the human genome, the development of high-throughput methods such as DNA microarrays
offers new insights into disease mechanisms, target identification,
and function as a possible diagnostic tool.
Currently, this technology is based on the analysis of an array,
per centimeter squared, of up to 100,000 oligonucleotides or
10,000 polymerase chain reaction (PCR) products. However, in
cancer studies only a small number of genes or expressed sequence tags are differentially expressed. Many bioinformatic
tools under development and testing are quite complex and/or
rely on these significantly expressed genes to establish a diagnosis
for unknown samples. The concept of typical tissue-specific genes
implies the existence of a few meaningful genes (tumor markers)
for diagnosis of NSCLC. The minimal number of predictor genes
for NSCLC is not known.
We have therefore generated a DNA microarray–based list of
differentially expressed genes and identified lung- or lung tumor–
specific target genes. We then tested the feasibility of using ratios
of gene expression levels and chosen thresholds to accurately
discriminate between lung tumor and normal lung tissue. These
diagnostic marker genes should allow reproducible tissue determination and an extension to routine clinical application.
METHODS
Tumor Samples
Tissue specimens from tumor and noninvolved lung of consecutive
patients with NSCLC (adenocarcinoma and squamous cell carcinoma)
or lung metastases, who underwent pulmonary resection surgery between 1999 and 2001, were included. One hundred ninety-nine snapfrozen lung tumors (n ⫽ 133), lung metastases (n ⫽ 8), and normal lung
tissues (n ⫽ 58) were used to create two data sets. Fifty-six squamous cell
carcinomas, 43 adenocarcinomas, and 15 noninvolved lung samples
were investigated by microarray analysis (Data Set A). Results of the
microarray analysis were proven by reverse transcription (RT)-PCR on
a second independent data set (Data Set B) consisting of samples from
17 squamous cell carcinomas, 17 adenocarcinomas, 8 lung metastases of
different primary tumors, and the corresponding noninvolved paired lung
tissues. Tumor histology was classified according to the World Health
Organization classification system (4).
The use of human tissues was approved by the local ethics committee
and the patients gave informed consent.
Microarray Expression Analysis
A total of 10 ␮g of RNA from each sample was used to prepare biotinylated target cRNA as previously described (5–7). A detailed protocol is available at www.affymetrix.com. Samples were hybridized to
a custom expression monitoring DNA microarray designed by Eos
Biotechnology (South San Francisco, CA), using Affymetrix GeneChip
technology (Affymetrix, Santa Clara, CA) (8), that contained essentially
all expressed human genes in the public domain at the time of design
(EOS-K). Sequences included on the array were derived from human
genomic, expressed mRNA, and expressed sequence tag databases in
GenBank (9). Consensus sequences representing human expressed sequences were generated with Clustering and Alignment Tools software
(DoubleTwist, Oakland, CA), and prediction of the expressed genome
from the human genome sequence was done by ab initio exon prediction
(10). The 59,000 probe sets on this microarray represent about 45,000
mRNA and expressed sequence tag clusters and 6,200 predicted exons.
Data were used after ␥ distribution normalization.
Gene Chip Analysis
(Received in original form January 29, 2004; accepted in final form June 1, 2004)
Supported by grants from EOS Biotechnology, DFG SFB 610 TPB1, SFB 598 TP
A5, Deutsche Krebshilfe (70-2787-Bu 3), and BMBF (01-ZZ0109). The sponsors
of the study had no role in data interpretation or writing of the report.
Correspondence and requests for reprints should be addressed to Hans-Stefan
Hofmann, M.D., Department of Cardiothoracic Surgery, Martin Luther Universität
Halle-Wittenberg, Ernst Grube Strasse 4, 06097 Halle, Germany. E-mail: stefan.
[email protected]
Am J Respir Crit Care Med Vol 170. pp 516–519, 2004
Originally Published in Press as DOI: 10.1164/rccm.200401-127OC on June 7, 2004
Internet address: www.atsjournals.org
For analysis of gene expression total RNA was extracted with TRIzol
(GIBCO, Karlsruhe, Germany) and biotinylated cRNA was prepared
by in vitro transcription after synthesis of double-stranded cDNA by
standard protocols. After cRNA fragmentation and hybridization with
microarrays (EOS-K chips), signals were detected with streptavidin–
phycoerythrin. Signal enhancement was performed with biotinylated
goat anti-streptavidin antibodies. Arrays were washed and stained with
the GeneChip Fluidics Station 400 (Affymetrix) and scanned with a
GeneArray scanner (Agilent Technologies, Palo Alto, CA). Primary
image analysis was performed by using Microarray Suite 5.0 (Affymetrix). Images were scaled to an average hybridization intensity of 200.
All expression values below 60 were set to 60.
Hofmann, Hansen, Burdach, et al.: Lung Neoplasm Discrimination
To identify specific genes that were differentially expressed in tumors as compared with normal lung tissue we used a criterion that
marks differential gene expression at an approximate significance level
(determined by the Bonferroni method) of 8.0 ⫻ 10⫺7, using a Student
t test, and a fold change cutoff of 2.0 (for upregulated genes) or 0.5
(for downregulated genes). Calculation of fold change was performed
by dividing the mean expression level of a gene in tumor samples by
the mean expression level of the same gene in normal lung samples.
RT-PCR
In a reverse transcription reaction, cDNA was synthesized from 200 ng
of total RNA with 100 units of SuperScript II reverse transcriptase (Invitrogen, Karlsruhe, Germany). A one-fifth volume of cDNA reaction
was used for semiquantitative PCR of RAGE and cyclin-B2 per 18S rRNA
amplification for intersample correction. Each PCR contained 1.5 mM
MgCl2, gene-specific primers (5 pmol each), dNTPs (10 ␮M each), and
1 unit of recombinant Taq DNA polymerase (Promega, Mannheim,
Germany). PCR primers were 5⬘-TGA ACA CAG GCC GGA CAG
AAG-3⬘ (sense) and 5⬘-CCC ATC CAA GTG CCA GCT AAG-3⬘
(antisense) for RAGE, 5⬘-AGC TGC TTC CTG CTT GTC TC-3⬘
(sense) and 5⬘-GCA CAA TGA AGC ACA CAT CC-3⬘ (antisense)
for cyclin-B2, and 5⬘-GTT GGT GGA GCG ATT TGT CTG G-3⬘
(sense) and 5⬘-AGGGCAGGGACTTAATCAACGC-3⬘ (antisense)
for 18S rRNA. PCR was performed after initial denaturation at 95⬚C
for 2 minutes and 30 seconds at 95⬚C, 20 seconds at 58⬚C for primer
annealing, and 40 seconds at 72⬚C, with 35 PCR cycles for RAGE, 37
cycles for cyclin-B2, and 12 cycles for 18S rRNA. After gel electrophoretic separation, the intensity of PCR products was densitometrically
evaluated with AIDA 2.0 software (Raytest, Straubenhardt, Germany).
In a first step the PCR analyses were optimized with a few samples to
obtain the same expression ratio as when using the DNA chip assay.
For internal signal correction between different PCR processes we
always used loading samples. Subsequently, the ratio between RAGE
and cyclin-B2 amplification was calculated.
Statistical Analysis
Results of gene expression are given as medians. Box-and-whisker plots
were constructed for illustration: the boundary of the box closest to
zero indicates the 25th percentile centered about the mean, the line
within the box marks the 50th percentile (median), and the boundary
of the box farthest from zero indicates the 75th percentile. Whiskers
517
above and below the box represent the 5th and 95th percentiles, respectively, and circles represent the 1st and 99th percentiles. Statistical
significance was determined on the basis of patient expression data,
using a Student t test.
RESULTS
Comparison of expression profiles determined by microarray
analysis (Data Set A) of normal lung and NSCLC samples revealed significant changes in gene expression in a total of 344
genes (0.6%). We detected 0.3% upregulated and 0.3% downregulated genes in tumor compared with normal lung tissue.
Among these differentially expressed genes we searched for
stable diagnostic targets.
The classic marker cytokeratin (CK-6A) is specific for squamous cell carcinomas, whereas thyroid transcription factor-1
(TTF-1) is specific for adenocarcinomas. These two markers can
be used to differentiate squamous cell carcinoma (gene chip
expression for CK-6A, 851.6 ⫾ 383.8 versus adenocarcinoma,
117.0 ⫾ 146.1) and adenocarcinoma (TTF-1, 379.0 ⫾ 203.4 versus
squamous cell carcinoma, 94.5 ⫾ 66.7). However, neither marker
can be used to entirely differentiate NSCLC from normal lung
tissue. In contrast, we found that the inverse expression, between
normal lung and NSCLC, of the genes encoding RAGE (receptor for advanced glycation end products) and cyclin-B2 was
highly diagnostic for nearly all NSCLC specimens (Figure 1).
RAGE is involved in the process of arteriosclerosis, diabetes,
aging, and Alzheimer’s disease (11). Cyclin-B2 is a member of
the cyclin family and plays an essential role in the cell cycle
machinery governing the transition from G2 to M phase (12).
The expression of RAGE mRNA alone was, in terms of median
values, fourfold downregulated in NSCLC, whereas cyclin-B2
mRNA was twofold upregulated (Figure 2). Using these two
genes identified in data set A, we calculated the expression ratio
per sample by dividing the expression value of RAGE by that
of cyclin-B2 (Figure 3A). On the basis of the median mRNA
ratio of RAGE to cyclin-B2, samples with ratio values greater
than 2 were identified as normal lung and those with ratio values
Figure 1. Gene expression for RAGE and cyclin-B2 in all normal
lung, adenocarinoma (AC), and squamous cell carcinoma (SC)
samples (Data Set A).
518
AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 170 2004
Figure 2. Box-and-whisker plots of gene
expression of RAGE (A ) and cyclin-B2 (B )
for the samples of Data Set A. p Values
of AC and SC versus normal lung group
were determined by t test.
less than 2 were identified as lung tumor. The sensitivity for
discrimination between NSCLC and normal lung tissue was determined as 87%, with a specificity of 100%.
Thereafter, we tested the accuracy of the RAGE:cyclin-B2
ratio in Data Set B. For this purpose we blinded all lung tumors,
metastases, and corresponding normal lung samples (n ⫽ 84)
and performed RT-PCR for expression analysis (Figure 4). We
confirmed that the ratio of RAGE per cyclin-B2 correctly distinguishes between NSCLC and normal lung with 94.7% accuracy.
All NSCLC tissues were classified correctly (sensitivity, 100%),
and 4 of the 42 normal lung tissues were incorrectly defined
(specificity, 90%).
The quotients of the overall RAGE and cyclin-B2 mRNA
levels for lung cancer and normal lung tissue determined by RTPCR (Data Set B) were comparable to the results of Data Set
A from the gene chip analysis (Figure 3B). Moreover, the expression ratio of RAGE mRNA per cyclin-B2 mRNA also correlated
positively in lung metastases with the ratios in adenocarcinomas
and squamous cell carcinomas. Also, all blinded lung metastasis
samples were correctly classified as tumor tissue.
DISCUSSION
Current gene expression profiling based on bioinformatics tools
is highly accurate in the diagnosis and classification of cancer.
The possibility of cancer classification based solely on gene expression monitoring by microarray analysis was first shown for
human acute leukemias (6). The translation of gene expression
data to potentially useful targets for molecular diagnosis depends
largely on statistical analysis. Although some statistical tests may
identify robust diagnostic target genes (13), the expression of
these genes is often extremely variable from sample to sample.
The expression ratio technique of two differentially expressed
genes can reduce this interindividual variability. The physiologic
function of these genes can be connected with tumorigenesis,
but plays no role in tissue discrimination by the mathematical
technique of expression ratios. We evaluated the ratio of RAGE
per cyclin-B2 as a reliable marker to differentiate normal lung
from NSCLC or lung metastases of different histologic origin.
Whereas the physiological function of cyclin-B2 seems to be
clear as a regulator of the cell cycle (12), the functional role of
RAGE in cancer needs to be evaluated. RAGE was previously
detected in a differential display experiment to be downregulated in lung cancer in comparison with corresponding normal
tissue (14). Data from the group of Huttunen showed a decreased
invasive transendothelial migration in vitro, using human fibrosarcoma cells, and suppression of lung metastasis formation
in vivo after treatment with the RAGE ligand amphoterin (15).
Glioma cells overexpressing RAGE had increased tumor growth
(16). These conflicting data may be explained by differences in
the experimental design as well as by organ-specific differences.
From our results it cannot be determined whether downregulation of RAGE is a consequence or cause of tumorigenesis. In
the latter case, we would predict a higher rate of lung tumors
in RAGE knockout mice, which has indeed been identified (A.
Bierhaus, personal communication).
The expression ratio technique represents an effective method
to translate the strengths of microarray expression profiling into
Figure 3. Box-and-whisker plots of the
RAGE/cyclin-B2 expression quotient determined by DNA microarray analysis (A ) for
normal lung, adenocarcinomas (AC), and
squamous cell carcinomas (SC)—Data Set
A. (B ) RAGE/cyclin-B2 quotient based on
RT-PCR in adenocarcinomas (AC), squamous cell carcinoma (SC), and lung metastases compared with paired normal lung
samples—Data Set B.
Hofmann, Hansen, Burdach, et al.: Lung Neoplasm Discrimination
519
manuscript; R.-E.S. does not have a financial relationship with a commercial entity
that has an interest in the subject of this manuscript; A.S. does not have a financial
relationship with a commercial entity that has an interest in the subject of this
manuscript.
References
Figure 4. Correlation of the blinded differentiation between tumor tissue (adenocarcinomas, squamous cell carcinomas, and lung metastases)
and paired normal lung samples and representative agarose gel electrophoresis after RT-PCR of amplified cDNA fragments for RAGE and cyclinB2 for two different patient samples of each tissue type.
simple clinical tests. The technique is simple and effective with
broad clinical use in diagnosis as well as prediction of prognosis
in cancer. This ratio-based technique is independent from the
expression measuring method, needs no gene expression reference (housekeeping gene as loading control), requires only small
pieces of RNA, and does not require the coupling of transcription
to translation for chosen genes (17). Gordon and coworkers
were the first to test expression ratios in the discrimination of
two different tissues (18). Using two or three expression ratios
of two differentially expressed genes, they found that the differential diagnosis of mesothelioma and pulmonary adenocarcinoma was 95 and 99% accurate, respectively. With the geometric
mean of the three most accurate individual ratios of four genes,
Bueno and coworkers could distinguish, with 90% accuracy,
normal prostate and prostate cancer samples obtained at surgery
(19). This technique is not only highly precise in the discrimination of cancer tissues; it can be equally useful in additional clinical
applications such as prediction of outcome. In patients with
mesothelioma, treatment-related outcome was predicted by gene
expression ratio-based analysis (17).
By using ratios of gene expression and rationally chosen
thresholds we have demonstrated an alternative and simple approach to predict clinical parameters such as malignancy of tissue
samples. This technique can be easily adapted and extended
to routine clinical application without the need for additional
sophisticated analyses.
Diagnoses provided by the expression ratio-based technique
on the basis of limited amounts of tissue (obtained by fine needle
aspiration, pleural effusion), without traditional histology results, have been shown to correlate with response to chemotherapy, and hence may be useful in predicting clinical course (20).
Our results, which demonstrate not only a simple characterization of NSCLC, could also be a first step in an automated diagnostic process. Further investigations will test the usefulness of the
RAGE/cyclin-B2 quotient for differentiation of lung neoplasms
from normal lung, and of other tumor types from their paired
normal tissue samples.
Conflict of Interest Statement : H.-S.H. does not have a financial relationship with
a commercial entity that has an interest in the subject of this manuscript; G.H.
does not have a financial relationship with a commercial entity that has an interest
in the subject of this manuscript; S.B. states that the results of this study do not
impact on the past grant as the present shares; B.B. does not have a financial
relationship with a commercial entity that has an interest in the subject of this
1. Parkin DM, Pisani P, Ferlay J. Estimates of the worldwide incidence of
eighteen major cancers in 1985. Int J Cancer 1993;54:594–606.
2. Greenlee RT, Hill-Harmon MB, Murray T, Thun M. Cancer statistics,
2001. CA Cancer J Clin 2001;51:15–36.
3. Liotta L, Petricoin E. Molecular profiling of human cancer. Nat Rev Genet
2000;1:48–56.
4. Travis WD, Colby TV, Corrin B, Shimosato Y, Brambilla E, editors.
Histological typing of lung and pleural tumours, 3rd ed. Berlin:
Springer-Verlag; 1999. p. 31–47.
5. Wodicka L, Dong H, Mittmann M, Ho MH, Lockhart DJ. Genome-wide
expression monitoring in Saccharomyces cerevisiae. Nat Biotechnol
1997;15:1359–1367.
6. Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov
JP, Coller H, Loh ML, Downing JR, Caligiuri MA, et al. Molecular
classification of cancer: class discovery and class prediction by gene
expression monitoring. Science 1999;286:531–537.
7. Tamayo P, Slonim D, Mesirov J, Zhu Q, Kitareewan S, Dmitrovsky E,
Lander ES, Golub TR. Interpreting patterns of gene expression with
self-organizing maps: methods and application to hematopoietic differentiation. Proc Natl Acad Sci USA 1999;96:2907–2912.
8. Lipshutz RJ, Fodor SP, Gingeras TR, Lockhart DJ. High density synthetic
oligonucleotide arrays. Nat Genet 1999;21(1 Suppl):20–24.
9. Benson DA, Karsch-Mizrachi I, Lipman DJ, Ostell J, Rapp BA, Wheeler
DL. GenBank. Nucleic Acids Res 2000;28:15–18.
10. Salamov AA, Solovyev VV. Ab initio gene finding in Drosophila genomic
DNA. Genome Res 2000;10:516–522.
11. Thornalley PJ. Cell activation by glycated proteins: AGE receptors, receptor recognition factors and functional classification of AGEs. Cell
Mol Biol (Noisy-le-grand) 1998;44:1013–1023.
12. Roberts JM. Evolving ideas about cyclins. Cell 1999;98:129–132.
13. Boer JM, Huber WK, Sultmann H, Wilmer F, von Heydebreck A, Haas
S, Korn B, Gunawan B, Vente A, Fuzesi L, et al. Identification and
classification of differentially expressed genes in renal cell carcinoma
by expression profiling on a global human 31,500-element cDNA array.
Genome Res 2001;11:1861–1870.
14. Schraml P, Bendik I, Ludwig CU. Differential messenger RNA and
protein expression of the receptor for advanced glycosylated end products in normal lung and non–small cell lung carcinoma. Cancer Res
1997;57:3669–3671.
15. Huttunen HJ, Fages C, Kuja-Panula J, Ridley AJ, Rauvala H. Receptor
for advanced glycation end products-binding COOH-terminal motif
of amphoterin inhibits invasive migration and metastasis. Cancer Res
2002;62:4805–4811.
16. Taguchi A, Blood DC, del Toro G, Canet A, Lee DC, Qu W, Tanji N,
Lu Y, Lalla E, Fu C, et al. Blockade of RAGE–amphoterin signalling
suppresses tumour growth and metastases. Nature 2000;405:354–360.
17. Gordon GJ, Jensen RV, Hsiao LL, Gullans SR, Blumenstock JE, Richards WG, Jaklitsch MT, Sugarbaker DJ, Bueno R. Using gene expressio ratios to predict outcome among patients with mesothelioma.
J Natl Cancer Inst 2003;95:598–605.
18. Gordon GJ, Jensen RV, Hsiao LL, Gullans SR, Blumenstock JE, Ramaswamy S, Richards WG, Sugarbaker DJ, Bueno R. Translation of microarray data into clinically relevant cancer diagnostic tests using gene
expression ratios in lung cancer and mesothelioma. Cancer Res 2002;
62:4963–4967.
19. Bueno R, Loughlin KR, Powell MH, Gordon GJ. A diagnostic test for
prostate cancer from gene expression profiling data. J Urol 2004;171:
903–906.
20. Sotiriou C, Powles TJ, Dowsett M, Jazaeri AA, Feldman AL, Assersohn
L, Gadisetti C, Libutti SK, Liu ET. Gene expression profiles derived
from fine needle aspiration correlate with response to systemic chemotherapy in breast cancer. Breast Cancer Res 2002;4:R3.